The present single-center, prospective observational cohort study was conducted between December 2022 and January 2023. Subjects diagnosed with COVID-19 who had complete lung CT data and blood test results were included. Subjects with the following conditions were excluded from the study: other infectious pneumonias, severe mental illness, atrial fibrillation, a history of myocardial infarction, severe renal insufficiency, severe valvular heart disease, and pregnancy. Additionally, individuals unable to obtain an AVI due to upper limb infections, previous vascular interventions, or amputations were excluded. All patients provided informed consent, and approval from the ethics committee (2024TG001) was obtained. The study adhered to the guidelines of the Declaration of Helsinki.

Patients were stratified into three groups—mild, moderate, and severe—according to the “Diagnosis and Treatment Protocol for Novel Coronavirus Pneumonia (Trial Version 10)” by the National Health Commission of China (19). Mild: This group includes individuals with upper respiratory tract infections, presenting symptoms such as dry throat, sore throat, cough, and fever. Moderate: Patients in this group have a persistent high fever for more than 3 days and/or experience symptoms such as cough and shortness of breath; however, their respiratory rate (RR) remains below 30 breaths/min, and their oxygen saturation is above 93% while resting on room air. Imaging shows a characteristic COVID-19 infection with pneumonia. Severe: Adults with any of the following that cannot be explained by other reasons besides the new coronavirus infection are included: (1) shortness of breath with an RR ≥ 30 beats/min, (2) oxygen saturation ≤ 93% on air inhalation at rest, (3) partial pressure of arterial oxygen (PaO2)/oxygen uptake (FiO2) ≤ 300 mmHg (1 mmHg = 0.133 kPa), with correction for PaO2/FiO2 according to the following formula in high altitude areas8 (more than 1,000 m): PaO2/FiO2 × [760/atmospheric pressure (mmHg)]. The PaO2/FiO2 should be corrected according to the following formula: PaO2/FiO2 × [760/atmospheric pressure (mmHg)]. (4) There is a progressive exacerbation of clinical symptoms, with a marked increase in the size of the lung lesion by >50% within 24–48 h on imaging. These criteria are used to categorize patients into mild, moderate, and severe groups.

All patients underwent a physical examination, and basic data were recorded after admission. Gender, age, smoking, height, weight, medical history, hypertension, diabetes, coronary artery disease, cerebrovascular disease, chronic renal disease, and pulse oxygen saturation (SpO₂) were recorded. Body mass index (BMI) was calculated as weight (kg)/height (m²).

Laboratory findings included leukocyte count, neutrophil percentage, lymphocyte percentage, C-reactive protein (CRP), and interleukin 6 (IL-6). The neutrophil-to-lymphocyte ratio (N/L ratio) was defined as the ratio of the neutrophil count to the lymphocyte count.

To exclude superadded bacterial infections, all enrolled patients underwent sputum testing and blood cultures within 48 h of hospital admission. Patients with clinically significant bacterial growth were excluded from this study.

Finally, the study also recorded the length of hospital stay and 30-day mortality rates.

AVI measurements were performed on all patients on the same day as their laboratory examinations. The AVI was measured by the cuff oscillation method using PASESA AVE-2000Pro (PASESA AVE-2000, DAIWA Healthcare, Shenzhen, China). Patients rested for at least 5 min before measurement, and avoided smoking and caffeinated beverages for at least 24 h before the measurement. The AVI, systolic blood pressure (SBP), diastolic blood pressure (DBP), and heart rate (HR) were measured with the patient in a seated position with a cuff around the left upper arm.

CT examinations were performed on all patients on the same day after the AVI measurements. Non-contrast-enhanced CT examinations were acquired using a multidetector CT scanner with 64 detector rows (uCT 760, United Imaging, Shanghai, China). The scanning parameters are as follows: tube voltage of 120 kV, automatic milliamp tube current, collimation width of 40 mm, screw pitch of 0.984, rotation time of 0.6 s, layer thickness of 5 mm, and matrix of 512 × 512. The patients were scanned in the supine position, and the scanning began after deep inspiration and breath-holding, without the use of contrast. The scan covered the area from the apex to the base of the lung.

An AI imaging diagnostic system (uAI Discover—Pneumonia, United Imaging Intelligence, Shanghai, China) was used for the automatic identification and quantitative analysis of infected lesions. A previous study showed that the system achieved high accuracy in automatically delineating lesion regions (with an average dice similarity coefficient of 91.6%) and lesion percentage metrics (with a sensitivity of 99.76% and a misdiagnosis rate of only 0.08%) compared to 95.3% for delineation (20).

The AI imaging system first segmented the network based on manually contoured CT data to create an initial model. Then, the initial model was applied to the next batch of infection areas, manual corrections were made, and the VB-Net model was finally built (21). A chest radiologist (with 10 years of experience) independently evaluated the CT images of all subjects without any knowledge of the patients’ signs, symptoms, or prognosis. The software automatically extracted images of the lungs (5 lung lobes and 18 lung segments) and lesion areas and automatically calculated each parameter. Parameters such as the whole lung and maximum lobe or segmental infection volume, infection percentage (the percentage of infection volume in the lung lobe segment), and the number of infected lobes or segments were obtained (Figure 1). Ground-glass opacity (GGO) is defined as a density distribution within the lesion ranging from −750 HU to −300 HU (Hounsfield units) on the CT images.

The primary outcome based on medical records is the 30-day mortality rate.

A total of 24 patients were randomly selected to assess repeatability. The AVI and the percentage of infection in the highest affected lobe percentages were measured by two doctors for inter-observer variability. One day later, one of the doctors remeasured the AVI and the percentage of infection in the highest affected lobe for intra-observer variability in these 24 patients.

Continuous data were described as the means and standard deviations and were compared using an ANOVA test. Non-normal variables were described as the medians and quartiles and compared using the Mann–Whitney U test. Categorical variables were described as numbers (percentages) and compared using a chi-squared test.

Because of the skewed distribution, the WBC count was log transformed to obtain normality. Spearman’s rank correlation coefficient was used to analyze the correlation between the AVI and laboratory parameters. Partial correlation was used to analyze the correlation between the AVI and age. Multiple backward linear regression analyses were used to investigate the independent associations between the AVI and inflammatory markers and radiographic parameters in patients in the moderate and severe groups, as the increase in AVI was more pronounced in these groups. The variables included in the analysis, whole lung, highest lobe, highest segmental of infection percentage, CRP, IL-6, BMI, age, and smoking status.

To study the fitted hazard ratios (HRs) and account for any confounding variables, a stepwise multivariable Cox proportional hazards analysis was conducted. An AVI ≥ 33 was considered high arterial stiffness (22), and a SpO₂ of ≤93% was considered low oxygen saturation (19). The following covariates were identified as most influential ones: age, male gender, heart rate, SpO₂ ≤ 93, N/L ratio, CRP > 8 mg/L, AVI ≥ 32, whole lung infection percentage, and highest lobe infection percentage. A log-rank test was conducted to determine the statistical significance of the Kaplan–Meier survival curves.

Bland–Altman plots were used to compare the AVI and highest lobe infection percentage measurements of the 24 subjects as mentioned above. The intraclass correlation coefficient (ICC) was used to evaluate inter- and intra-observer variabilities. All statistical analyses were performed using SPSS 23.0 (IBM, Armonk, NY, USA) statistical software. A two-tailed p-value of less than 0.05 was considered statistically significant.

A total of 100 subjects, 44 men and 56 women, aged from 18 to 80 years, with a mean age of 76 years, were included. Table 1 lists the basic data of patients and includes their clinical and laboratory characteristics.

No significant differences were observed in smoking status, BMI, SBP, DBP, and history of hypertension among the three groups (all p > 0.05). Leukocyte count, neutrophil percentage, lymphocyte percentage, N/L ratio, CRP, and IL-6 levels were higher in the severe group than in the mild group (p < 0.05). The AVI in the moderate and severe groups was significantly higher than in the mild group [mild: 15.13 ± 3.93 (min 9; max 23) vs. moderate: 21.79 ± 5.00 (min 14; max 31) vs. severe: 21.06 ± 6.45 (min 10; max 46), p < 0.001].

In both the moderate and severe groups, the infection volume of the whole lung, the highest lobe, the highest segment, the GGO, the left lung, and the right lung, along with the infection percentage of the whole lung, the highest lobe, the GGO, the left lung, the right lung, and the number of infected lobes and lobe segments were higher than in the mild group (p < 0.05). No significant difference was observed between the moderate and severe groups (Table 2, Figure 2).

In unadjusted analyses, the AVI was positively correlated with age and the number of infected lobes in all patients (r = 0.262, 0.026, respectively; all p < 0.05). After adjusting for age, this association changed such that positive correlations were observed between the AVI and the number of infected lobes and infected segments (r = 0.012, 0.016, p < 0.05, respectively) (Table 3). In a multiple linear regression involving moderate–severe patients, the AVI was significantly associated with the highest lobe infection percentage (β = 0.813; 95%CI, 0.056–0.394, p = 0.011) (Table 4).

The results of the Cox regression are shown in Figure 3. In the univariable Cox regression model, age, the N/L ratio, and an AVI of ≥33 remained positively associated with incident death. No significant impact was observed for heart rate, SpO₂ ≤ 93%, CRP>8 mg/L, whole lung infection percentage, or highest lobe infection percentage. In the multivariable Cox proportional hazards analysis, an AVI ≥ 33 was independently associated with incident death (HR, 16.201; 95% CI: 1.601, 163.987; p < 0.001).

The results of the Kaplan–Meier curve are shown in Figure 4. Survival was longer for patients with an AVI < 33 compared to patients with an AVI ≥ 33 (HR, 16.201; 95% CI: 1.601, 163.987; p < 0.001). The log-rank test indicated that this difference was statistically significant (p < 0.05).

Bland–Altman agreement plots for the AVI and highest lobe infection percentages are shown in Figure 5 and Table 5. The ICCs for the AVI for inter-and intra-observer reproducibility were 0.902 (95% CI, 0.787–0.956; p < 0.001) and 0.896 (95% CI, 0.720–0.941; p < 0.001), respectively. The ICCs for highest lobe infection percentages for inter-and intra-observer reproducibility were 1.000 (95% CI, 0.999–1.000; p < 0.001) and 1.000 (95% CI, 0.999–1.000; p < 0.001), respectively.